Skip to main content
Click to return to IBM ECVG home


Evaluating Authentication Systems Using Bootstrap Confidence Intervals
R. Bolle, N. Ratha, and S. Pankanti
Proc. 1999 IEEE Workshop on Automatic Identification Advanced Technologies (WAIAT-99), Morristown NJ, pp. 9-13, October 1999.

Reporting the matching performance of biometrics identification or authentication systems is a controversial issue and perhaps an issue that is not well understood. It seems apparent that False Reject Rates and False Accept Rates are the measurements to express system accuracy. The size of the database of biometrics signals which is used to obtain estimates of the error rates is another important parameter that should be reported along with the error estimates. This database size can be used to compute confidence intervals of the error rates. Further, some indication of the quality of the fingerprint impressions should also be given. This paper develops bootstrap confidence intervals for estimating the accuracy of the error estimates and bootstrap techniques for indicating the quality and diversity of the prints in the database. The size of the database is also related to the accuracy of the error estimates. Results of our experiments on a sample fingerprint database is reported.

PDF version


  Privacy | Legal | Contact | IBM Home | Research Home | Project List | Research Sites | Page Contact